急诊与创伤影像学人工智能:ASER AI/ML 专家组 Delphi 共识声明——研究指南、实践与优先事项
该共识由 ASER AI/ML 专家组于 2022–2024 年开展 Delphi 共识流程,第一阶段由 7 名指导委员会成员确立数据整理、有效性、人为因素、工作流程、障碍、未来方向与伦理等关键主题并生成声明长清单,第二阶段进行两轮匿名 RAND/UCLA Likert 评分(第一轮网络数据采集,第二轮带文献超链接的定制 Excel 文档),轮间通过编辑与知识综合最大化共识,≥80% 同意率的声明纳入最终文件。第一轮和第二轮分别含 81 和 78 个条目,21 名专家中 18 名(86%)回复第一轮、15 名回复第二轮(脱落 17%),最终 65 条声明达成共识。一致共识集中在透明的方法学报告、用外部数据测试泛化性与稳健性、以合适指标和基线进行性能基准测试。该文件旨在为急诊与创伤影像 AI 研究者提供促进最佳实践和进一步讨论的框架。
要点
- 共识由 ASER AI/ML 专家组于 2022–2024 年通过两轮 Delphi(RAND/UCLA Likert 评分)形成,≥80% 同意率的声明纳入最终文件。
- 最终 65 条声明达成共识,第一轮 18/21 名专家(86%)回复、第二轮 15 名回复(脱落 17%)。
- 一致共识的核心主题包括:透明的方法学报告、用外部数据测试泛化性与稳健性、以合适指标和基线进行性能基准测试。
- 共识覆盖数据整理、有效性、人为因素、工作流程、障碍、未来方向与伦理等主题,作为急诊/创伤影像 AI 研究的框架文件。
适用人群与场景
- 从事急诊与创伤影像人工智能研究的研究者,涵盖从数据整理和算法开发到上市后监测与再训练的技术成熟度各阶段。
诊断标准与关键阈值
- Delphi 声明纳入标准:≥80% 同意率。
研究设计与关键结果
- 研究类型:ASER AI/ML 专家组 Delphi 共识,2022–2024 年开展;第一阶段 7 名指导委员会成员确立关键主题并生成编辑整理的声明长清单;第二阶段进行两轮匿名 RAND/UCLA Likert 评分(第一轮网络数据采集,第二轮带文献超链接的定制 Excel 文档),轮间编辑与知识综合以最大化共识。
- 第一轮和第二轮分别含 81 和 78 个条目;18/21 名专家(86%)回复第一轮,15 名回复第二轮(脱落 17%);最终 65 条声明达成共识。
- 一致共识集中在透明的方法学报告、用外部数据测试泛化性与稳健性、以合适指标和基线进行性能基准测试。
证据类型、等级与局限
- 证据类型:专家共识(Delphi 共识声明);采用匿名 RAND/UCLA Likert 评分,≥80% 同意率纳入声明。
- 局限:原文未报告局限性。
对床旁实践的改变
- 该文件作为框架,旨在促进急诊与创伤影像 AI 研究者之间的最佳实践和进一步讨论。
指南或共识出处
- ASER AI/ML 专家组 Delphi 共识声明《Artificial intelligence in emergency and trauma radiology: ASER AI/ML expert panel Delphi consensus statement on research guidelines, practices, and priorities.》(Emerg Radiol)。
提炼说明
- 本条依据论文摘要整理;摘要未涉及「处置流程与药物剂量」「红旗征与不能漏的情况」,故未列出这些栏目;全文请点「阅读原文」。
英文原文(摘要)· 点击展开
BACKGROUND: Emergency/trauma radiology artificial intelligence (AI) is maturing along all stages of technology readiness, with research and development (R&D) ranging from data curation and algorithm development to post-market monitoring and retraining.
PURPOSE: To develop an expert consensus document on best research practices and methodological priorities for emergency/trauma radiology AI.
METHODS: A Delphi consensus exercise was conducted by the ASER AI/ML expert panel between 2022-2024. In phase 1, a steering committee (7 panelists) established key themes- curation; validity; human factors; workflow; barriers; future avenues; and ethics- and generated an edited, collated long-list of statements. In phase 2, two Delphi rounds using anonymous RAND/UCLA Likert grading were conducted with web-based data capture (round 1) and a bespoke excel document with literature hyperlinks (round 2). Between rounds, editing and knowledge synthesis helped maximize consensus. Statements reaching ≥80% agreement were included in the final document.
RESULTS: Delphi rounds 1 and 2 consisted of 81 and 78 items, respectively.18/21 expert panelists (86%) responded to round 1, and 15 to round 2 (17% drop-out). Consensus was reached for 65 statements. Observations were summarized and contextualized. Statements with unanimous consensus centered around transparent methodologic reporting; testing for generalizability and robustness with external data; and benchmarking performance with appropriate metrics and baselines. A manuscript draft was circulated to panelists for editing and final approval.
CONCLUSIONS: The document is meant as a framework to foster best-practices and further discussion among researchers working on various aspects of emergency and trauma radiology AI.
参考来源
- Dreizin D, Khatri G, Staziaki PV, et al. Artificial intelligence in emergency and trauma radiology: ASER AI/ML expert panel Delphi consensus statement on research guidelines, practices, and priorities. Emerg Radiol. 2024. PMID: 39714735. doi:10.1007/s10140-024-02306-1
供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。